AMD Moves All 7nm CPU, GPU Production to TSMC

Earlier this year, we suggested that AMD’s decision to move its 7nm GPU production to TSMC could be a sign of trouble for GF’s 7nm ramp. Today, AMD is announcing that it will move all of its 7nm production on both CPUs and GPUs to TSMC. We already knew that TSMC was AMD’s foundry of choice for its 7nm Vega GPU, and the company had later announced that its 7nm Epyc CPUs (codenamed Rome) would be built there as well, but this is still a significant announcement for the company.

AMD’s Mark Papermaster has published a new blog post in which he writes:

AMD’s next major milestone is the introduction of our upcoming 7nm product portfolio, including the initial products with our second generation “Zen2” CPU core and our new “Navi” GPU architecture. We have already taped out multiple 7nm products at TSMC, including our first 7nm GPU planned to launch later this year and our first 7nm server CPU that we plan to launch in 2019. Our work with TSMC on their 7nm node has gone very well and we have seen excellent results from early silicon. To streamline our development and align our investments closely with each of our foundry partner’s investments, today we are announcing we intend to focus the breadth of our 7nm product portfolio on TSMC’s industry-leading 7nm process. We also continue to have a broad partnership with GLOBALFOUNDRIES spanning multiple process nodes and technologies. We will leverage the additional investments GLOBALFOUNDRIES’ is making in their robust 14nm and 12nm technologies at their New York fab to support the ongoing ramp of our AMD Ryzen, AMD Radeon, and AMD EPYC processors. We do not expect any changes to our product roadmaps as a result of these changes.

This news is not entirely surprising. When we visited GlobalFoundries in February, the company’s now-CEO Thomas Caulfield, noted that the foundry was making major investments in its 22FDX and 12FDX nodes and intended to be a market leader in those segments, but that it had settled on a “fast follower” strategy for 7nm. 22FDX and 12FDX are FD-SOI nodes that are attracting attention from the automotive, IoT, RF, and other sectors that don’t benefit from the design characters of FinFETs but would still like a node transition that reduced idle power and offered better overall characteristics. At the same time, however, GF was clearly moving ahead with 7nm and EUV installations — we saw the enormous ASML EUV installation underway and the subfloor of the fab was under active construction to support the tool requirements.

The most likely explanation for this shift is a difference in business alignments. When AMD spun off GlobalFoundries, the entire point was to create a foundry partner that would be able to compete with TSMC and Intel at the leading edge. GF would benefit from new sources of investment and revenue that weren’t tied to AMD and would benefit from additional customers for its factories. Nine years after the spinoff, things simply haven’t played out that way. The two companies tangled badly in 2011 – 2012 with poor Llano yields and an inability to ramp Krishna and Wichita that left AMD no choice but to move those two CPUs to TSMC, where they became Kabini and Temash.

Things didn’t improve at 14nm. GlobalFoundries initially planned to skip 20nm altogether in favor of its own version of 14nm, dubbed 14XM. The company was later forced to cancel 14XM and instead licensed 14nm technology from Samsung, becoming a second source foundry on that node. That deal appears to have worked well — Ryzen has certainly proven to be strong competition for Intel, and the 12nm optimization that led to second-generation Ryzen this year has also been effective. But GlobalFoundries appears to have stumbled again when it comes to its own 7nm. Skipping 10nm was supposed to give it a boost in terms of time-to-market, but now the company’s largest publicly announced customer has taken all of its designs elsewhere. GF is still supposed to have a 7nm node ready for high volume manufacturing by the end of the year, but who is going to use it?

AMD says this won’t impact any of its roadmaps and we’re inclined to take the company at its word. This shift has been hinted at in reports for some time and it’s clear the decision wasn’t made spur of the moment. The only puzzle at this point is when, exactly, we should look for 7nm Ryzen chips. AMD hasn’t yet announced tapeout of Ryzen 2, which may imply that it plans to lead with Rome on 7nm rather than a consumer launch. The company has used both strategies in the past — the first Opterons launched ahead of consumer Athlon 64s, while the original K10 silicon was available as Barcelona in servers before shipping as Phenom for consumers. More recently, chips like Ryzen led in consumer before debuting as part of Epyc. In the modern era, server chips have usually taken longer than consumer counterparts due to the higher difficulty of validating high core count processors, but AMD’s modular architecture allows them to simplify that step in some regards (designing the interconnects on a 32-core CPU is still no trivial task, regardless of whether the building blocks are simpler).

AMD, of course, has its own reasons to want to push ahead aggressively on 7nm. With Intel having slipped on 10nm and now forecasting availability only in Q4 2019, AMD has a chance to steal a march on Intel and be the first x86 company to debut a new process node. AMD effectively led for a time on 180nm when both it and Intel introduced new CPUs within a month of each other. Intel launched the Pentium III Coppermine in October of 1999 and AMD followed with the Athlon K7.5 (Pluto core) in November of the same year. In that case, however, AMD’s yields were far superior to Intel’s at higher clock speeds due to the Athlon’s reliance on L2 cache that ran at a fraction of CPU clock, while Intel’s Coppermine L2 was clocked at full CPU speed. AMD decisively outshipped Intel in total 1GHz CPUs at the time, while Intel was eventually forced to recall its Pentium III 1.13GHz CPU for overheating. Advantage at 180nm? AMD.

But never again. While the company didn’t slip behind all at once, Intel gradually widened its process node leadership. At first, this wasn’t particularly problematic — 90nm + Prescott were disastrous for Intel — but eventually, the gap began to tell. By 2012, Intel was running a full node ahead of AMD. While TSMC’s 7nm node and Intel’s 10nm are expected to be roughly equivalent, AMD still has the chance to launch on 7nm before Intel (and will, of course, take advantage of the fact that people don’t necessarily know that there’s a gap in technical capability between 10nm and 7nm at different foundries).

Now, AMD wants to steal a march on Intel and presumably align with a foundry partner that’s got the capacity and manufacturing capability to churn out 7nm cores — and apparently TSMC is a much better fit for those goals than GlobalFoundries, which will remain a major AMD manufacturing partner on 12nm and 14nm.

Continue reading

Microsoft Has Asked AMD to Boost Xbox Series S, Series X Production

Microsoft's feeling the heat, in terms of console shipments — so much so, it's asked AMD directly for any help it can provide.

Report: Packaging Issues, PS5 Demand May Be Hurting TSMC Production

The hardware shortages currently hitting most of the PC market may be caused by a shortage in a necessary component in chip manufacturing, not low yields on TSMC's 7nm node.

TSMC Will Spend an Extra $10B on Chip Production in 2021

TMSC is ramping up its capital expenditures next year, investing an additional $10B into advanced chip manufacturing.

Report: Intel Will Outsource Core i3 Production to TSMC’s 5nm Node

This would be the first time the chip giant has built one of its Core CPUs on a different company's process node. It's not an exaggeration to call this a pivotal moment in Intel's history.